基于景象匹配的无人机侦察视频快速配准方法

摘要

In order to improve the effect and speed of Unmanned Aerial Vehicle (UAV) reconnaissance video registration,a fast registration method based on scene matching is proposed.Firstly,a feature matching algorithm based on AGAST-Difference and Fast Retina Keypoint (FREAK) is proposed to match the video frames.Then,the matching region search method is proposed to find the matching region for the target frame in the satellite digital map.Finally,the video frame is registered to matching regions,and the homography matrix is calculated out according to 2D relationship of the matching regions for implementing the splicing.The experimental results show that:1) The feature matching algorithm based on AGAST-Difference and FREAK has great advantages in scale,rotation,viewpoint and so on;and 2) The matching region search method can avoid the error caused by the Position and Orientation System (POS) and the introduction of control points,thus the correction accuracy and speed are improved.With the splicing speed up to 25 frames per second for videos of 810 × 612 resolution,the 2D localization accuracy of our algorithm for a ground target can reach 7.87 m when the UAV flies at about 1000 m above ground.%为了改善无人机侦察视频配准效果与速度,提出一种基于景象匹配的无人机侦察视频快速配准方法.首先提出基于AGAST-Difference与FREAK的特征匹配算法对视频帧之间配准,然后提出匹配区域搜索方法在数字卫星地图上找到视频帧匹配区域,最后将视频帧与匹配区域配准,根据匹配区域二维关系计算单应矩阵完成拼接.实验结果表明,基于AGAST-Difference与FREAK的特征匹配算法在尺度、旋转、视点等变化及运行速度上存在很大优势,匹配区域搜索方法避免了定位定向系统带来的误差与引入控制点,提高了纠正精度与速度.本配准方法对像素大小为810 ×612的视频拼接速度达到25帧/s,在离地约1000 m的空中,二维定位精度可达7.87 m.